AI Red Teaming — Adversarial ML Security Assessment

Comprehensive AI red team assessment conducted at SGP IMMO in collaboration with Google's AI Red Teamer program. Built three representative ML models (spam classifier, network anomaly detector, malware image classifier) and systematically tested them against adversarial attacks including prompt injection (direct/indirect), label flipping data poisoning, model reverse engineering, and insecure output handling (XSS). 20+ attack simulations executed with measurable impact documented in a 90-page report.